Message-ID: <9e60cc0d-43d1-4110-9299-cc7348fd8295@arm.com>
Date: Fri, 16 Jan 2026 09:51:19 +0000
From: Ben Horgan <ben.horgan@....com>
To: Reinette Chatre <reinette.chatre@...el.com>, linux-kernel@...r.kernel.org
Cc: tony.luck@...el.com, Dave.Martin@....com, james.morse@....com,
 babu.moger@....com, bp@...en8.de
Subject: Re: [PATCH v1] fs/resctrl: Add missing kconfig entry for
 CONFIG_RESCTRL_ASSIGN_FIXED

Hi Reinette,

On 1/15/26 17:19, Reinette Chatre wrote:
> Hi Ben,
> 
> On 1/15/26 7:46 AM, Ben Horgan wrote:
>> Hi Reinette,
>>
>> On 1/15/26 15:37, Reinette Chatre wrote:
>>> Hi Ben,
>>>
>>> On 1/15/26 2:32 AM, Ben Horgan wrote:
>>>> On 1/14/26 19:09, Reinette Chatre wrote:
>>>>> On 1/13/26 6:49 AM, Ben Horgan wrote:
>>>>>> The commit 3b497c3f4f04 ("fs/resctrl: Introduce the interface to display
>>>>>> monitoring modes") introduced CONFIG_RESCTRL_ASSIGN_FIXED but did not add
>>>>>> the kconfig entry. Add this.
>>>>>>
>>>>>> Also, take the opportunity ensure that any user attempt to change the
>>>>>> assign mode fails from the resctrl code rather than delegating to the arch
>>>>>> specific code and let the user know by adding a message in last_cmd_status.
>>>>>
>>>>> Why *any* attempt? Avoiding delegating to arch seems ok as a goal but this change
>>>>> additionally changes interface with user space. Current behavior when user writes
>>>>> existing mode to the file is to just return success. For example, if current mode
>>>>> is "default" and user writes "default". This patch changes this behavior to fail
>>>>> instead. Is this intended?
>>>>
>>>> Intended but as you point out, it would be best to accept the current
>>>> value without error for maximum compatibility between architectures.
>>>> I'll update the config option help text to reflect this. Was the help
>>>> text also discussed previously?
>>>
>>> It is not clear to me which help text you are referring to. Could you
>>> please point me to it?
>>
>> I just mean the description in Kconfig. Maybe I'm using the wrong
>> terminology?
> 
> Counter assignment mode is the right terminology and used throughout the resctrl
> documentation.

Ack.

> 
>>
>> From this patch:
>>     help
>>      Enabled by the architecture when the counter assignment mode is not
>>      configurable. This ensures that counter assignment is not
> 
> nit: I recommend above "counter assignment" be "counter assignment mode" to be specific
> that this is about the *mode* being configurable and not talking about the actual counter
> assignment.
> 
> Also please ensure that the formatting adheres to the guidance in "Kconfig configuration
> files" in Documentation/process/coding-style.rst. Specifically, "help text is indented an
> additional two spaces".

Will do.

> 
>> advertised as
>>      configurable and configuration attempts fail.
> 
> "configuration attempts" -> "attempts to change counter assignment mode"?

Sure.

> 
> As a sidenote, if I understand correctly, MPAM can actually support changing the mode
> when it is in "free running" mode (when there are enough counters/monitors available). It
> seems to me these systems *could* advertise something like:
> 
> 	# cat /sys/fs/resctrl/info/L3_MON/mbm_assign_mode
> 	[default]
> 	mbm_event
> 
> It is not clear to me if there is a scenario that would benefit from such capability though.
> I just bring this up to confirm whether CONFIG_RESCTRL_ASSIGN_FIXED is something that
> MPAM enabling wants to enforce for all platforms?

On a platform with sufficient monitors to have one per rmid idx
(num_partid * num_pmg) it would be architecturally possible to use
either mode but the policy decision is that the driver will always
decides which mode to use at probe based on whether  there are
sufficient monitors or not. Hence, we want to enforce
CONFIG_RESCTRL_ASSIGN_FIXED for all platforms.

> 
> Reinette

Thanks,

Ben

Powered by blists - more mailing lists